Impact of Photovoltaic Penetration on Small Signal
Stability considering Uncertainties
Samundra Gurung, Sumate Naetiladdanon, Anawach Sangswang
Department of Electrical Engineering
King Mongkut’s University of Technology, Thonburi
Bangkok, Thailand
Samundra24@gmail.com
Abstract—This paper applies cumulant method to analyze the
influence of Photovoltaic Penetration (PV) considering
fluctuations in small signal stability The cumulant method for
small signal stability is mainly based on sensitivity analysis
which is then used with Gram-Charlier expansion to obtain the
valuable information of probability density function (PDF) and
Cumulative Density Function (CDF) of the real part of the
critical eigen value and damping factor. A supplementary
controller to enhance small signal stability is also presented and
its performance is also analyzed. The test system is a modified
IEEE 16 machine, 68 bus system which is benchmarked to study
small signal oscillatory dynamics in power systems. The results
show that the PV fluctuations have potential to cause oscillatory
instability and the future large PV farms should also contain
Power Oscillation Damper (POD) to damp these oscillations.
Index Terms—Cumulant Method, Power Oscillation Damper,
Probabilistic Analysis, Probability Density Function
I. INTRODUCTION
Recently, there is a large proliferation of Photovoltaic
Power Generation (PVG) in electric power system and their
aggregated production capacity is rapidly approaching
towards the conventional generation capacity. However,
Photovoltaic Power greatly suffers from intermittency which
could severely hamper the power system. Moreover, PV
resources in nearby places could be correlated due to
insolation, temperature and other environmental factors and
can cause different electrical issues in power system [1].
There has been a great effort in understanding their effects in
power system stability especially small signal stability.
However, most of the works are based on deterministic
analysis [2-3] or are considered only for small penetration in
distribution level [4].
The PV power fluctuations can be studied by using
probabilistic method. This method characterizes uncertainties
by describing the variation by suitable probability
distribution. Probabilistic Method gives system planners
better idea about future system conditions which helps them
to make well decision to operate power system in a reliable
manner [5].
978-1-5386-4950-3/17/$31.00 ©2017 IEEE
Broadly speaking, the current probabilistic method to
analyze small signal stability can be categorized into
analytical and non-analytical method. The analytical method
is based on the assumption that the system eigen values are a
function of some uncertain input parameters. These methods
employ complex mathematical approximations to infer the
probability distribution of important electrical variables such
as eigen values and damping factor [5].
The non-analytical method uses Monte Carlo Simulation
[MCS] technique to obtain the information about system
eigen value distribution. This technique is a numerical
method which needs running of multiple deterministic small
signal stability analysis [5]. This leads to huge computation
burden and many researchers only use it to comparison
purpose.
This paper uses an analytical method based on cumulants
to analyze the effect of PV generation in small signal
stability. The cumulant method was chosen as it gives better
results when compared with other analytical methods such as
Point Estimate Method or Probability Collocation Method
[6]. The output of the cumulant method is used with Gram-
Charlier Expansion technique to obtain the Probability
density function (PDF) and Cumulative density function
(CDF) of real part of critical eigen value and its damping
factor. The results obtained for different cases are all
compared with the MCS output. Moreover, a Power
Oscillation Damper (POD) is designed to damp the
oscillation and its performance is also seen.
The paper is organized as follows: Section II gives
theoretical background on some essential ideas such as PV
converter and POD modeling as well as the probabilistic
method. Section III discusses about the PV irradiance
modeling. Section IV contains the results of the analysis for
different cases which is followed by conclusion in section V.
II. NECESSARY THEORETICAL BACKGROUND
A. POD Model
The Power oscillation damping model is shown in Fig.1.
The POD provides a supplementary signal in the PV
controller which helps to damp the oscillation. It has a similar
structure to a Power system stabilizer (PSS). The input to system damping. This paper uses reactive power modulation
POD can be any signals like bus voltage, line current, active technique as the system damping increases more with more
or reactive power etc. The POD has a washout block with a real power injection in this scheme [9].
gain K and Time constant Tw. This output is then given to The whole model was built in DIGSILENT.
lead lag compensator which has time constants T1, T2, T3 and C. Probabilistic Small Signal Stability
T4. The output is then fed as a modulating signal to the
It is generally considered that the Beta distribution is the
converter reactive power controlling loop as described before.
most applicable distribution to describe the PV irradiance
fluctuation [10]
The probability density function (pdf) of the solar irradiance
can be written as:
1
f (S ) = S a -1 (1- S )b -1 (1)
Figure 1. POD Model B ( a, b )
B. PV Model Where, S is the irradiance, a and b are the shape
parameters and B is the beta function.
The PV model is developed as suggested by NERC and The power output of the PV generation varies linearly
WECC and is similar to the grid side converter model of wind with the irradiance [1] and thus, the PDF of the power output
turbine generators (Type 4 WTG) [7]. The model considers (fp) can be derived using the well-known transformation
PV as a Voltage Source behind internal impedance and is method as:
connected to the grid as shown in Fig. 2. a −1 b −1
The PV source injects real (p) and reactive power (q) to 1 1  P   P 
f p ( P) =   1 −  for 0 < P ≤ Pmax (2)
the grid via transformer which is modeled as impedance with Prated B(a, b)  Prated   Prated 
resistance (r) and reactance (x). The magnitude of point of 0 else
common coupling (PCC) voltage is denoted by VT and the
phase angle is ϕv. Where, P is the PV power injection, Prated is the rated PV
output power and Pmax is the maximum output power.
The PV power fluctuation is defined as ΔPi = Pi - Pd, where
Pd is the deterministic PV power generation.
The raw moment of the PV power variation can be
computed as follows:
Prated

α 
(n)
Figure 2. PV Farm connected to the Grid
ΔPi = x n f p ( x)dx
0
Fig.3. shows the converter model of the PVG. The voltage (3)
source can be decomposed into direct axis voltage (Vd) and n
(− Pi )n − k
quadrature axis voltage (Vq) which is supplied by two delay =  Cn k
* B (k + a, b)
k =0 B (a, b)
blocks representing the converter delay (TV) [8]. This portion
of the converter is in synchronously rotating grid reference n!
, B ( k + a, b) = τ ( k + a )τ (b)
k
frame. The controller part represented by PI controller is in Where, Cn =
terminal voltage oriented frame where the real part of the (k !(n − k )!) τ (k + a + b)
current controls the active power and the imaginary part of and τ is the gamma function.
the current controls reactive power. The active and reactive The nth order cumulants can be calculated as follows:
power is controlled from reference real power (Pref) and
reference reactive power (Qref) respectively. k (1) ΔPi = α (1) ΔPi
k (2) ΔPi = α (2) ΔPi − α (1) ΔP i
(4)
k (3) ΔPi = α (3) ΔPi − 3α (1) ΔPi α (2) ΔPi + 2(α (1) ΔPi )3
.........
Assuming there are N PV system in the power system and
λk = σk + jωk, which denotes the kth eigen value (critical) of
Figure 3. Model of PV controller the power system with real part (σ) and imaginary part (ω),
the following relationship can be established [11],
The output (iqmod) from POD described in the last heading
is connected to the reactive control loop to enhance the
 ∂λk N
Δλk = Δσ k + j Δωk =  ΔPi (5)
i =1 ∂Pi

∂λk λk ( Pi + ΔPi ) − λk ( Pi )
= (6)
∂Pi ΔPi
The output cumulants (knΔσ) can be calculated from the
input cumulants using (7). It is then used to obtain the output
central moments which is used to find the PDF and CDF of
the critical eigen value with the help of Gram-Charlier
Expansion [5].
n Figure 4. PDF of the Solar Irradiance
  ∂λ  
N
k Δσ (=   Re  k   k ( n ) ΔPi
(n)
(7)
The accuracy of the proposed probabilistic power model
i =1   ∂Pi   is calculated by using average root mean square (ARMS) [1]
It is easy to derive similar relationship for the damping N
factor (ζ) of critical eigen value using (8),  (F M o d ,i − F R e f ,i ) 2
−σ ARMS = i =1 (10)
ξ= (8) N
σ 2 +ω2 Where, FMod,i and FRef,i are the ith value on the CDF
curves of the fitting model and the reference respectively and
Finally, the following two terms are calculated, N is the selected number of points. The ARMS for this case is
0 0.242% showing the validity of the proposed probabilistic
P(σ < 0) = Fσ (0) = 
−∞
fσ ( x)dx model.
(9) IV. CASE STUDY
0
Fig. 5 shows the test system which has 16-machines and
P(ξ < 0) = Fξ (0) = 
−∞
fξ ( x)dx 5 areas and is a reduced order equivalent of the
interconnected New England Test System (NETS) and New
Where, Fσ and Fζ are the CDF of critical eigen value real York power system (NYPS). The test system has three PV
part and critical eigen value damping factor respectively and generators connected at buses 18, 41 and 42. The value of the
fσ and fζ are the PDF of the critical eigen value real part and parameter of PV farm is given in the appendix. All the PV
critical eigen value damping factor respectively. Generators are operated at 0.83pu for both case A and B.
III. PROBABILISTIC MODEL OF PV GENERATION The test system has all Power system stabilizers (PSS)
disabled and the excitation system is modeled as a simple fast
The actual daily data collected from 2013 to 2016 by CES
acting static exciter whose value is given in appendix [13].
Solar Cells Testing Center (CSSC), Thonburi is used to
construct the PDF of the PV irradiance. The hourly data
between 12pm to 1 pm for the month of May was used to
construct the histogram shown in Fig. 4.
The histogram clearly shows a bimodal distribution and
was thus, fitted separately to the best possible distribution by
using Matlab statistics tool box.
Only the higher irradiance portion was considered as it
contained large frequency of data and we might also
encounter errors if we consider the total distribution [12]. The
best fit was found to be beta distribution and the shape
parameters were obtained using maximum log likelihood test
whose values are given in Table I.
TABLE I
PARAMETERS OF SOLAR IRRADIANCE Figure 5. Modified IEEE 68 Bus System [14]

a b Mean Variance Base A. PV penetration without POD

(pu) Irradiance
In this first case, the deterministic small signal stability
(W/m2)
analysis was run and the critical eigen value was found to be
12.2101 2.48237 0.831045 0.00894751 1100 -0.00915779+j*0.00429. The cumulant method was then used
to obtain the CDF and PDF of real part of critical eigen value
and damping factor.
The PDF of the real part of critical eigen value is shown
in Fig.6. which clearly has some part of its area lying in the
positive half suggesting the possibility of instability.
Fig.7 shows the CDF of the real part of the critical eigen
value and can be used to calculate the following:
0
P(σ < 0) = Fσ (0) =
−∞
 fσ ( x)dx = 80.94%

Figure 8. PDF of the damping factor of the Critical Eigen Value

This shows that the given system has 19.06% of being
stochastically unstable due to PV fluctuations even though it
was deterministically stable.
For comparison, the histogram of MCS is also shown. We
can see that the histogram of MCS matches the Gram-
Charlier expansion approximations.

Figure 9. CDF of the Real part of the Critical Eigen Value

Figure 6. PDF of the Real part of the Critical Eigen Value
B. PV penetration with POD
The second case considers PV farms which have POD
installed in it. The input signal for PODs of PV farm 1 and 2
is taken as current from Bus 18 to 42 and current from 41 to
42 for POD of PV farm 3. The main reason for selecting this
signal was they are local and thus, more readily available.
The POD parameters were tuned extensively to obtain the
best possible values which are given in the appendix. Fig.10
shows the PDF of the critical eigen value. Unlike in the
previous case, the whole area of PDF always lies in the
negative plane suggesting the system to be small signal
stable.

Figure 7. CDF of the Real part of the Critical Eigen Value

The PDF of the damping factor of critical eigen value is

shown in Fig. 8 which has some part of its area lying on the
negative left hand side suggesting possible instability. The
CDF of the damping factor of critical eigen value is shown in
Fig.9 and can be used to find:
∞
P(ξ > 0) = Fξ (0) =  fξ ( x)dx = 80.4% Figure 10. PDF of the Real part of the Critical Eigen Value with POD
0
This result shows that there is a probability of 19.6% that Fig. 11 shows the CDF of the real part of the critical
the damping factor of critical eigen value can be negative eigen value and can be used to find:
which can lead the system to be oscillatory instability. 0
P(σ < 0) = Fσ (0) = 
−∞
fσ ( x)dx = 100%

PV Rating
(MW)
100
Figure 11. CDF of the Real part of the Critical Eigen Value with POD
This results shows that the system which has PV
Generators with POD installed is always small signal stable.
C. PV penetration at different level
Finally, all PV generators without POD is varied from 10
to 100 percent (0.1 to 1 pu) of its value and the probability of
damping factor of critical eigen value being less than zero [1]
was calculated. Fig. 12 shows that the probability of
instability increases with the penetration level and is highest
when all the PV farms operate at their rated value. [2]
The probability of damping factor of critical eigen being
less than zero is always nil percent for the case of PVG with
POD installed and is thus, not shown for comparison. [3]
[4]
[5]
[6]
Figure 12. Probability of Instability in percentage for different penetration
levels [7]
V. CONCLUSION
In this paper, the cumulant method is used to analyze the [8]
effect of PV generation in small signal stability. The output of
the cumulant method provides the important statistical
parameters which are used with Gram Charlier Expansion to [9]
obtain the PDF and CDF of the real part of the critical eigen
value and damping factor.
It was found that the system was stochastically unstable [10]
even though it was deterministically stable which showcase
the effect of PV irradiance fluctuations. The paper also
[11]
proposes one possible method to improve stability in such
case by including a supplementary damping controller in PV
control structure. It was found that it greatly enhances small
signal stability. Finally, the effect of PV penetration on [12]
different levels was considered and was seen that it higher
penetration leads to the deterioration of system stability.
[13]
[14]
APPENDIX
TABLE II
PARAMETERS OF THE PV CONTROLLER
Proportional gain Integral gain Converter time
delay
0.1 0.05 0.015
TABLE III
POD PARAMETERS
KW Tw T1 T2 T3 T4
1 10 2 0.2 0.3 0.2
TABLE IV
EXCITATION SYSTEM PARAMETERS
K Ta Vmax Vmin
38 0.1 10 -10
REFERENCES
M. Fan, V. Vittal, G. T. Heydt and R. Ayyanar, "Probabilistic Power
Flow Studies for Transmission Systems With Photovoltaic Generation
using Cumulants,” IEEE Trans. Power Systems, vol. 27, pp. 2251-2261,
Nov. 2012.
S. Gurung, S. Naetiladdanon and A. Sangswang, "Impact of Large
Photovoltaic Penetration on small signal stability," in IEEE Region 10
Conference (TENCON)., pp. 646-650.
S. Eftekharnejad, V. Vittal, G. T. Heydt, B. Keel and J. Loehr "Small
Signal Stability Assessment of Power Systems with Increased
Penetration of Photovoltaic Generation: A Case Study," IEEE Trans.
Sustainable Energy, vol. 4, pp. 960-967, Oct. 2013.
S. Liu, P. X. Liu and X. Wang, "Stochastic Small-Signal Stability
Analysis of Grid-Connected Photovoltaic Systems," IEEE Trans.
Industrial Electronics, vol. 63, pp. 1027-1037, Feb. 2016.
J. L. Reuda, D. G. Colome and I. Erlich, "Assessment and
Enhancement of Small Signal Stability Considering Uncertainties,"
IEEE Trans. Power Systems, vol. 24, pp. 198-207, Feb. 2009.
R. Preece, K. Huang and J. V. Milanovic, "Probabilistic Small-
Disturbance Stability Assessment of Uncertain Power Systems Using
Efficient Estimation Methods," IEEE Trans. Power Systems, vol. 29,
pp. 2509-2517, Sept. 2014.
B. Tamimi, C. Canizares and K. Bhattacharya, "System Stability
Impact of Large-Scale and Distributed Solar Photovoltaic Generation:
The Case of Ontario, Canada," IEEE Trans. Sustainable Energy, vol. 4,
pp. 680-688, July 2013.
J. Fortmann, S. Engelhardt, J. Kretschmann, C. Feltes and I. Erlich
"New Generic Model of DFG-Based Wind Turbines for RMS-Type
Simulation," IEEE Trans. Energy Conversion, vol. 29, pp. 110-118,
March 2014.
S. Y. Ruan, G. J. Li, B. T. Ooi and Y. Z. Sun "Power system damping
from real and reactive power modulations of voltage-source-converter
station," IET Generation, Transmission and Distribution, vol. 2, pp.
311-320, 2008.
S. Rahman, M. A. khallat and Z. M. Salameh "Characterization of
Insolation Data for Use in Photovoltaic System Analysis Models,"
Energy, vol. 13, pp. 63-72, 1988.
S. Q. Bu, W. Du, H.F.Wang, Z.Chen, L.Y.Xiao and H. F. Li
"Probabilistic Analysis of Small-Signal Stability of Large- Scale Power
Systems as Affected by Penetration of Wind Generation," IEEE Trans.
Power Systems, vol. 27, pp. 762-770, May 2012.
Y. Zhou, Y. Li, W. Liu, Z. Li and J. Liu "The Stochastic Response
Surface Method for Small-signal Stability Study of Power System with
Probabilistic Uncertainties in Correlated Photovoltaic and Loads,"
IEEE Trans. Power Systems, in press.
P. Kundur, “Power System Stability and Control,” Tata Mc-Graw Hill,
New Delhi, 2006, p.814.
C. Canizares et. al, "Benchmark Models for the Analysis and Control of
Small-Signal Oscillatory Dynamics in Power Systems," IEEE Trans.
Power Systems, in press.